Hi Ulf (and mailing-list),
I've an SD-card on a STM32MP157F-DK2 board that cannot switch to
high-speed mode:
"mmc0: Problem switching card into high-speed mode!"
On this board, it is not possible to switch to UHS modes.
And there is no power cycle done in kernel.
When checking the differences when I add full-pwr-cycle in DT, I see
that the OCR we ask the card is different:
0x300000 (MMC_VDD_32_33 | MMC_VDD_33_34) vs 0x200000 (MMC_VDD_33_34).
If I add this missing MMC_VDD_32_33 voltage range (without power cycle),
then the card can switch to high-speed.
Checking where this is done in the framework, I've seen something that
could correct my issue in mmc_select_voltage():
diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index 368f10405e13..bcd8fa81f78b 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -1132,7 +1132,7 @@ u32 mmc_select_voltage(struct mmc_host *host, u32 ocr)
mmc_power_cycle(host, ocr);
} else {
bit = fls(ocr) - 1;
- ocr &= 3 << bit;
+ ocr &= 3 << (bit - 1);
if (bit != host->ios.vdd)
dev_warn(mmc_dev(host), "exceeding card's
volts\n");
}
The ocr given to mmc_select_voltage() is 0x300000.
fls(ocr) = 22, bit = 21, 3 << bit = 0x600000.
With the &= operator, we then have only 0x200000 and have removed
MMC_VDD_32_33 mode.
The architecture is an Armv7, I hope that the fls() has the same
behavior on other architectures.
But as this function is also used for eMMC and SDIO, this could have
impacts I've not seen.
Maybe the issue is just with this SD-card, that doesn't properly handle
the range MMC_VDD_33_34 alone, and it could be out of spec.
I then have 3 possibilities:
- stop using this type of card if it is out-of-specs
- add full-pwr-cycle in this board's DT, but I'll have issues with other
boards that really cannot do power cycle
- push the proposed patch in mmc_select_voltage()
What's your opinion?
Best regards,
Yann
